88 lines
2.8 KiB
Python
88 lines
2.8 KiB
Python
"""冲突严重程度分级纯函数。
|
||
|
||
依据：
|
||
- 高（high）：相似度 >= HIGH_CONFLICT_THRESHOLD → 制度文件核心条款直接冲突
|
||
- 中（medium）：相似度 >= MEDIUM_CONFLICT_THRESHOLD → 程序/用词层歧义
|
||
- 低（low）：其它（如命名相似但实质无关）
|
||
"""
|
||
|
||
from __future__ import annotations
|
||
|
||
from typing import Tuple
|
||
|
||
|
||
def classify_conflict_severity(
|
||
similarity: float,
|
||
thresholds: Tuple[float, float] = (0.85, 0.92),
|
||
) -> str:
|
||
"""根据条款相似度返回冲突等级。
|
||
|
||
thresholds = (medium_min, high_min)，须 0 <= medium_min <= high_min <= 1。
|
||
返回：'high' / 'medium' / 'low'。
|
||
"""
|
||
medium_min, high_min = thresholds
|
||
if similarity >= high_min:
|
||
return "high"
|
||
if similarity >= medium_min:
|
||
return "medium"
|
||
return "low"
|
||
|
||
|
||
def normalize_policy_text(s: str) -> str:
|
||
"""规范化制度文本：去空白 + 中文/英文标点统一。
|
||
|
||
用于语义比对前的预处理，确保章节编号差异不干扰相似度计算。
|
||
转换规则：
|
||
- 全角空格 → 半角空格
|
||
- 全角 ASCII（数字 0-9 / 字母 A-Z a-z / 常见标点）→ 半角
|
||
- 合并连续空白为单空格
|
||
"""
|
||
if s is None:
|
||
return ""
|
||
# 全角 ASCII 标点 → 半角（覆盖冒号/逗号/句号/分号/问号/感叹号/括号/方括号/引号）
|
||
full_to_half_punct = {
|
||
"，": ",", "。": ".", "：": ":", "；": ";",
|
||
"？": "?", "！": "!", "（": "(", "）": ")",
|
||
"【": "[", "】": "]", "『": "'", "』": "'",
|
||
"「": "'", "」": "'",
|
||
}
|
||
out_chars = []
|
||
for ch in str(s):
|
||
code = ord(ch)
|
||
# 全角空格
|
||
if ch == " ":
|
||
out_chars.append(" ")
|
||
continue
|
||
# 已映射的标点
|
||
if ch in full_to_half_punct:
|
||
out_chars.append(full_to_half_punct[ch])
|
||
continue
|
||
# 全角数字 0xFF10-0xFF19 → 0x30-0x39
|
||
if 0xFF10 <= code <= 0xFF19:
|
||
out_chars.append(chr(code - 0xFEE0))
|
||
continue
|
||
# 全角字母 0xFF21-0xFF3A / 0xFF41-0xFF5A → 半角
|
||
if 0xFF21 <= code <= 0xFF3A:
|
||
out_chars.append(chr(code - 0xFEE0))
|
||
continue
|
||
if 0xFF41 <= code <= 0xFF5A:
|
||
out_chars.append(chr(code - 0xFEE0))
|
||
continue
|
||
out_chars.append(ch)
|
||
text = "".join(out_chars)
|
||
# 合并连续空白为单空格
|
||
return " ".join(text.split())
|
||
|
||
|
||
def parse_policy_draft_id(s: str) -> str:
|
||
"""解析送审稿 ID：去空白 + 校验非空。
|
||
|
||
占位骨架的轻量校验，业务层后续接入更严格规则。
|
||
"""
|
||
if s is None:
|
||
raise ValueError("policy_draft_id 不能为空")
|
||
s = str(s).strip()
|
||
if not s:
|
||
raise ValueError("policy_draft_id 不能为空")
|
||
return s
|